rollandf / network-operator / 25002584503

62%

master: 62%

Build Type

push

github

Committed by rollandf

Commit Message

fix: backfill ds-owner label on OFED pods when upgrading from GA releases

When upgrading from a GA release (e.g. v26.1) to v26.4+, existing MOFED
pods lack the ds-owner label that was added to the pod template in v26.4.
Because the DaemonSet uses OnDelete update strategy, these pods are never
recreated automatically and retain their original label set.

The upgrade controller calls BuildState with both nvidia.com/ofed-driver
and ds-owner label selectors. With no pods matching the combined selector,
BuildState errors with "driver DaemonSet should not have Unscheduled pods"
on every reconcile, preventing ApplyState from ever running and causing
the upgrade to stall indefinitely.

Add handleMissingDSOwnerLabelOnPods to the migration path. It runs at
operator startup before any controller is unblocked (MigrationCh gate),
finds OFED pods that lack ds-owner, and patches them with the correct
NicClusterPolicy CRD name. This is a no-op on v26.4+ upgrades where all
pods already carry the label.

Signed-off-by: Fred Rolland <frolland@nvidia.com>

Coverage Stats

25 of 35 new or added lines in 1 file covered. (71.43%)

95 existing lines in 9 files now uncovered.

4858 of 7844 relevant lines covered (61.93%)

46.24 hits per line